- Cuda toolkit v10.2 with a corresponding driver
nvcc
in your$PATH
- Properly set
$CUDA_HOME
Install them via poetry
:
poetry install
cd gpu_utils && python setup.py install
You can test these extensions by running pytest
in the gpu_utils
directory after
you've installed the extensions.
To run an experiment issue python fashion_mnist_experiment.py
. To configure it you
need to change global variables on the top of the file.
These experiments take a cuda device number as an argument, as well as sequence length.
Example usage: python rnn_addition_experiment.py -cuda 0 20
.